comparing SciPy IFFT against cuFFT

Please help,

I am trying to figure out the difference between Python SciPy and cuFFT with regards to forward and/or inverse 1d FFT. I don’t really have a coding question per say, more a general results question. I am new to SciPy FFT and IFFT and hope someone has some information.

I have a single input CSV file such that each line contains a single complex number of the form: “real, imaginary” I am reading into a cuFFT FFT call (complex-to-complex) and a SciPy FFT call. However I get wildly different results from cuFFT versus SciPy. I have done this same operation between cuFFT and fftwf for complex-to-complex but the comparisons are generally similar - some variation as I expect.

Can anyone think of a reason why SciPy would not at least somewhat match cuFFT complex-to-complex FFT (or Inverse FFT for that matter)? Is SciPy performing some kind of normalization or scaling by default?

Thank you for any help/hints

The CUFFT inverse FFT is is not normalized and generally needs to be normalized to match output from other tools such as Matlab. I’m not sure about SciPy.

The normalization is a common question on the web and is mentioned in the cufft manual.

Thanks for the reply. I will do some more digging on SciPy.

I was just wondering if anyone else has tried comparing SciPy IFFT/FFT against cuFFT IFFT/FFT for complex-to-complex data types and got very different result(s). It is strange that I can compare fftwf IFFT/FFT against cuFFT IFFT/FFT for complex-to-complex and the results match more or less.